Analysing the Potential Impact of Labeling Disagreements for Engineering Sensor Data.

摘要: We present the hyperbola recognition problem in Ground Penetrating Radar – GPR data as an example for pattern complex engineering sensor data. Traditionally, are analyzed manually by human experts a tedious and time-consuming process, e.g., to deduce positioning of linear object underneath roads just before reconstruction works take place. For supporting this process using Machine Learning methods, one needs have accurate ground truth derive models out it. As acquisition such annotated is impossible even quasi-ideal case, we 700 radargram images manually. This paper presents discusses outcomes study concludes, that single evaluation criteria compare performances GPR-focused methods might not be enough.